Abstract
The ability to represent multiple objects via a statistical summary is known as ensemble encoding. While humans can rapidly extract a summary feature from an ensemble of objects (e.g., mean orientation), representations of single objects may be considerably less accurate. Moreover, research has shown that single-item representations are biased towards the mean, which is consistent with the dominance of global over local processing (Navon, 1977). Despite this, the influence of local information on global processing is apparent in object-scene interactions (Lowe et al., 2015). Given the relationship between ensemble processing and scene perception (Cant & Xu, 2015), we investigated relationships between global and local processing utilizing a novel ensemble-interference paradigm. Participants were shown eight oriented triangles and instructed to remember their orientations. In a subsequent 2AFC task, a target and distractor were presented, and participants reported either the average orientation (global condition) or the orientation of a randomly selected single triangle (local condition). Additionally, we manipulated the degree of interference from the distractor. In the global condition, the distractor was either a non-target triangle from the studied set (high interference), or a novel triangle (low interference). Similarly, in the local condition, the distractor was either the mean of the set (high interference) or a novel triangle (low interference). When the mean was presented as the distractor in the local condition, participants were more likely to report it as the single target item, but were above chance at reporting the single item when the mean was not presented. Interestingly, we also observed local-to-global interference wherein single items from the set in the global condition biased representations of the mean. Thus, even though global features were reported more accurately and caused greater interference, we observed reciprocal interference between global and local processing, which is consistent with what is observed in object-scene interactions.
Meeting abstract presented at VSS 2018